Search

The CernVM File System (CVMFS) provides the software distribution backbone for High Energy and Nuclear Physics experiments and many other scientific communities in the form of a globally available shared software area. It has been designed for the software distribution problem of experiment software for LHC Runs 1 and 2. For LHC Run 3 and even more so for HL-LHC (Runs 4-6), the complexity of the experiment software stacks and their build pipelines is substantially larger. For instance, software is being distributed for several CPU architectures, often in the form of containers which includes base and operating system libraries, the number of external packages such as machine learning libraries has multiplied, and there is a shift from C++ to more Python-heavy software stacks that results in more and smaller files needing to be distributed. For CVMFS, the new software landscape means an order of magnitude increase of scale in several key metrics. This contribution reports on the performance and reliability engineering on the file system client to sustain current and expected future software access load. Concretely, the impact of the newly designed file system cache management is shown, including significant performance improvements for HEP-representative benchmark workloads, and an up to 25% performance increase in software built-time when the build tools reside on CVMFS. Operational improvements presented include better network failure handling, error reporting, and integration with container runtimes. And a pilot study using zstd as compression algorithm shows that it could bring significant improvements for remote data access times.

After using ROOT’s TTree I/O subsystem for over two decades and storing more than an exabyte of compressed High Energy Physics (HEP) data, advances in technology have motivated a complete redesign, RNTuple, which breaks backward-compatibility to take better advantage of these storage options. The RNTuple I/O subsystem has been designed to address performance bottlenecks and other shortcomings of TTree. Specifically, RNTuple comes with an updated, more compact binary data format that can be stored both in ROOT files and natively in object stores. It is designed for modern storage hardware (e.g. high-throughput low-latency NVMe SSDs), and provides robust and easy to use interfaces. The binary format of RNTuple is scheduled to become production grade in 2024, and recently has become mature enough to start exploring the integration into software used by HEP experiments. In this contribution, we discuss the developments to support the features as required by the ATLAS analysis Event Data Model (EDM) in RNTuple, which will enable its integration into the Athena software framework. With these developments in place, we evaluate the performance of the current most recent versions of RNTuple-based ATLAS data sets and compare this to that of TTree.

The RNTuple I/O subsystem is ROOT’s future event data file format and access API. It is driven by the expected data volume increase at upcoming HEP experiments, e.g. at the HL-LHC, and recent opportunities in the storage hardware and software landscape such as NVMe drives and distributed object stores. RNTuple is a redesign of the TTree binary format and API and has shown to deliver substantially faster data throughput and better data compression both compared to TTree and to industry standard formats. In order to let HENP computing workflows benefit from RNTuple’s superior performance, however, the I/O stack needs to connect efficiently to the rest of the ecosystem, from grid storage to (distributed) analysis frameworks to (multithreaded) experiment frameworks for reconstruction and ntuple derivation. With the RNTuple binary format soon arriving at its first production release, we present RNTuple’s feature set, integration efforts, and its performance impact on the time-to-solution. We show the latest performance figures of RDataFrame analysis code of realistic complexity, comparing RNTuple and TTree as data sources. We discuss RNTuple’s approach to functionality critical to the HENP I/O (such as multithreaded writes, fast data merging, schema evolution) and we provide an outlook on the road to its use in production.

RDataFrame is ROOT’s high-level interface for Python and C++ data analysis. Since it first became available, RDataFrame adoption has grown steadily and it is now poised to be a major component of analysis software pipelines for LHC Run 3 and beyond. Thanks to its design inspired by declarative programming principles, RDataFrame enables the development of highperformance, highly parallel analyses without requiring expert knowledge of multi-threading and I/O: user logic is expressed in terms of self-contained, small computation kernels tied together by a high-level API. This design completely decouples analysis logic from its actual execution, and opens several interesting avenues for workflow optimization. In particular, in this work we explore the benefits of moving internal data processing from an event-by-event to a bulkby-bulk loop. This refactoring dramatically reduces the framework’s runtime overheads; in collaboration with the I/O layer it improves data access patterns; it exposes information that optimizing compilers might use to auto-vectorize the invocation of user-defined computations; finally, while existing user-facing interfaces remain unaffected, it becomes possible to additionally offer interfaces that explicitly expose bulks of events, useful e.g. for the injection of GPU kernels into the analysis workflow. In order to inform similar future R&D, design challenges will be presented, as well as an investigation of the relevant timememory trade-off backed by novel performance benchmarks.

The recent evolutions of the analysis frameworks and physics data formats of the LHC experiments provide the opportunity of using central analysis facilities with a strong focus on interactivity and short turnaround times, to complement the more common distributed analysis on the Grid. In order to plan for such facilities, it is essential to know in detail the performance of the combination of a given analysis framework, of a specific analysis and of the installed computing and storage resources. This contribution describes performance studies performed at CERN, using the EOS disk-based storage, either directly or through an XCache instance, from both batch resources and highperformance compute nodes which could be used to build an analysis facility. A variety of benchmarks, both synthetic and based on real-world physics analyses and their corresponding input datasets, are utilized. In particular, the RNTuple format from the ROOT project is put to the test and compared to the latest version of the TTree format, and the impact of caches is assessed. In addition, we assessed the difference in performance between the use of storage system specific protocols, like XRootd, and FUSE. The results of this study are intended to be a valuable input in the design of analysis facilities, at CERN and elsewhere.

Particle physics has an ambitious and broad experimental programme for the coming decades. This programme requires large investments in detector hardware, either to build new facilities and experiments, or to upgrade existing ones. Similarly, it requires commensurate investment in the R&D of software to acquire, manage, process, and analyse the shear amounts of data to be recorded. In planning for the HL-LHC in particular, it is critical that all of the collaborating stakeholders agree on the software goals and priorities, and that the efforts complement each other. In this spirit, this white paper describes the R&D activities required to prepare for this software upgrade.

In the HEP community, software plays a central role in the operation of experiments’ facilities and for reconstruction jobs, with CVMFS being the service enabling the distribution of software at scale. In view of High Luminosity LHC, CVMFS developers investigated how to improve the publication workflow to support the most demanding use cases. This paper reports about recent CVMFS developments and infrastructural updates that enable faster publication into existing repositories. A new CVMFS component, the CVMFS Gateway, allows for concurrent transactions and the use of multiple publishers, increasing the overall publication rate on a single repository. Also, the repository data has been migrated to Ceph-based S3 object storage, which brings a relevant performance enhancement over the previously-used Cinder volumes. We demonstrate how recent improvements allow for faster publication of software releases in CVMFS repositories by focusing on the LHCb nightly builds use case, which is currently by far the most demanding one for the CVMFS infrastructure at CERN. The publication of nightly builds is characterized by a high churn rate, needs regular garbage collection, and requires the ability to ingest a huge amount of software files over a limited period of time.

Containers became the de-facto standard to package and distribute modern applications and their dependencies. The HEP community demonstrates an increasing interest in such technology, with scientists encapsulating their analysis workflow and code inside a container image. The analysis is first validated on a small dataset and minimal hardware resources to then run at scale on the massive computing capacity provided by the grid. The typical approach for distributing containers consists of pulling their image from a remote registry and extracting it on the node where the container runtime (e.g., Docker, Singularity) runs. This approach, however, does not easily scale to large images and thousands of nodes. CVMFS has long been used for the efficient distribution of software directory trees at a global scale. In order to extend its optimized caching and network utilization to the distribution of containers, CVMFS recently implemented a dedicated container image ingestion service together with container runtime integrations. CVMFS ingestion is based on per-file deduplication, instead of the per-layer deduplication adopted by traditional container registries. On the client-side, CVMFS implements on-demand fetching of the chunks required for the execution of the container instead of the whole image.

The CernVM File System provides the software and container distribution backbone for most High Energy and Nuclear Physics experiments. It is implemented as a file system in user-space (Fuse) module, which permits its execution without any elevated privileges. Yet, mounting the file system in the first place is handled by a privileged suid helper program that is installed by the Fuse package on most systems. The privileged nature of the mount system call is a serious hindrance to running CernVM-FS on opportunistic resource and supercomputers. Fortunately, recent developments in the Linux kernel and in the Fuse user-space libraries enabled fully unprivileged mounting for Fuse file systems (as of RHEL 8), or at least outsourcing the privileged mount system call to a custom, external process. This opens the door to several, very appealing new ways to use CernVM-FS, such as a generally usable “super pilot” consisting of the pilot code bundled with Singularity and CernVM-FS, or the on-demand instantiation of unprivileged, ephemeral containers to publish new CernVM-FS content from anywhere. In this contribution, we discuss the integration of these new Linux features with CernVM-FS and show some of its most promising, new applications.

The CernVM File System (CernVM-FS) provides a scalable and reliable software distribution service implemented as a POSIX read-only filesystem in user space (FUSE). It was originally developed at CERN to assist High Energy Physics (HEP) collaborations in deploying software on the worldwide distributed computing infrastructure for data processing applications. Files are stored remotely as content-addressed blocks on standard web servers and are retrieved and cached on-demand through outgoing HTTP connections only. Repository metadata is recorded in SQLite catalogs, which represent implicit Merkle treeencodings of the repository state. For writing, CernVM-FS follows a publish-subscribe pattern with a single source of new content that is propagated to a large number of readers. This paper focuses on the work to move the CernVM-FS architecturein the direction of a responsive data distribution system. A new distributed publication backend allows scaling out large publication tasks across multiple machines, reducing the time to publish. For the faster propagation of new published content, the addition of a notification system allows clients to subscribe to messages about changes in the repository and to request new root catalogs as soon as they become available. These devel-opments make CernVM-FS more responsive and are particularly relevant for use cases where a short propagation delay from repository down to individual clients is important, such as using CernVM-FS as an AFS replacement for distributing software stacks. Additionally, they permit the implementation of more complex workflows, with producer-consumer pipelines, as for example in the ALICE analysis trains system.

The CernVM File System (CernVM-FS) provides a scalable and reliable software distribution and—to some extent—a data distribution service. It gives POSIX access to more than a billion binary files of experiment application software stacks and operating system containers to end user devices, grids, clouds, and supercomputers. Increasingly, CernVM-FSalso provides access to certain classes of data, such as detector conditions data, genomics reference sets, or gravitational wave detector experiment data. For most of the high- energy physics experiments, an underlying HTTP content distribution infrastructure is jointly provided by universities and research institutes around the world. In this contribution, we will present recent developments and future plans. For future developments, we put a focus on evolving the content distribution infrastructure and at lowering the barrier for publishing into CernVM-FS. Through so-called serverless computing, we envision cloud hosted CernVM-FS repositories without the need to operate dedicated servers or virtual machines. An S3 compatible service in conjunction with a content delivery network takes on data provisioning, replication, and caching. A chainof time-limited and resource-limited functions (so called “lambda function” or “function-as- a-service”) operate on the repository and stage the updates. As a result, any CernVM-FS client should be able to turn intoawriter, possession of suitable keys provided. For repository owners, we aim at providing cost transparency and seamless scalability from very small to very large CernVM-FS installations.

LHC experiments make extensive use of web proxy caches, especially for software distribution via the CernVM File System and for conditions data via the Frontier Distributed Database Caching system. Since many jobs read the same data, cache hit rates are high and hence most of the traffic flows efficiently over Local Area Networks. However, it is not always possible to have local web caches, particularly for opportunistic cases where experiments have little control over site services. The Open High Throughput Computing (HTC) Content Delivery Network (CDN), openhtc.io, aims to address this by using web proxy caches from a commercial CDN provider. Cloudflare provides a simple interface for registering DNS aliases of any web server and does reverse proxy web caching on those aliases. The openhtc.io domain is hosted on Cloudflare's free tier CDN which has no bandwidth limit and makes use of data centers throughout the world, so the average performance for clients is much improved compared to reading from CERN or a Tier 1. The load on WLCG servers is also significantly reduced. WLCG Web Proxy Auto Discovery is used to select local web caches when they are available and otherwise select openhtc.io caching. This paper describes the Open HTC CDN in detail and provides initial results from its use for LHC@Home and USCMS opportunistic computing.

The performance gap between CPU and memory widens continuously. Choosing the best memory layout for each hardware architecture is increasingly important as more and more programs become memory bound. For portable codes that run across heterogeneous hardware architectures, the choice of the memory layout for data structures is ideally decoupled from the rest of a program. This can be accomplished via a zero‐runtime‐overhead abstraction layer, underneath which memory layouts can be freely exchanged. We present the low‐level abstraction of memory access (LLAMA), a C++ library that provides such a data structure abstraction layer with example implementations for multidimensional arrays of nested, structured data. LLAMA provides fully C++ compliant methods for defining and switching custom memory layouts for user‐defined data types. The library is extensible with third‐party allocators. Providing two close‐to‐life examples, we show that the LLAMA‐generated array of structs and struct of arrays layouts produce identical code with the same performance characteristics as manually written data structures. Integrations into the SPEC CPU® lbm benchmark and the particle‐in‐cell simulation PIConGPU demonstrate LLAMA's abilities in real‐world applications. LLAMA's layout‐aware copy routines can significantly speed up transfer and reshuffling of data between layouts compared with naive element‐wise copying. LLAMA provides a novel tool for the development of high‐performance C++ applications in a heterogeneous environment. [ABSTRACT FROM AUTHOR]